One of the ideas I took away from my conversation with Dan Greene of Radical Candor on Corey-osity Unleashed sits on my list as a rule. Praise should outnumber criticism five to one. Minimum.
You've heard some version of this. Every leadership workshop seems to have a slide with "5:1" on it.
I agree with the direction. I don't trust the number. Once you see where the number came from, you'll stop counting and start doing something more useful.
Where 5:1 Came From
Two separate research streams get mashed together into one slogan. Neither of them says what the slide says.
The marriage lab
John Gottman studied couples. According to the Gottman Institute, successful couples in conflict conversations spent five seconds in a positive or neutral emotional state for every one second in a negative state. Outside of conflict, the stable couples ran closer to 20:1.
Notice three things. These were married couples, not teams. The unit was seconds, not comments. And the measurement happened during arguments, not during a sprint review.
The business-team lab
The workplace version traces back to Marcial Losada. In the 1990s he ran a lab called the Capture Lab for the computer services firm EDS. Sixty business teams of eight people held meetings while observers behind one-way mirrors coded what they said.
Losada and Emily Heaphy reported the positive-to-negative ratio for each group. High-performing teams came in at 5.614. Medium teams at 1.855. Low performers at 0.363. Business writers rounded the top number to "about six" and the slogan took off. Jack Zenger and Joseph Folkman built a 2013 HBR piece, The Ideal Praise-to-Criticism Ratio, around the 5.6 figure.
Then it got stranger. Losada fed the numbers into the Lorenz equations, a set of differential equations from fluid dynamics used to model air convection. In 2005 he and Barbara Fredrickson published a paper in American Psychologist claiming a "critical" ratio of 2.9013. Above it, people flourish. Below it, they languish. Fredrickson wrote a popular book about the "3-to-1 ratio."
Four decimal places. For human emotion. I've shipped enough software to get nervous when a model produces a number with more precision than its inputs.
Then Somebody Checked the Math
Nick Brown, a graduate student in applied positive psychology, didn't buy it. He teamed up with physicist Alan Sokal and psychologist Harris Friedman. In 2013 they published The Complex Dynamics of Wishful Thinking in the same journal.
Their verdict: they found "no theoretical or empirical justification for the use of differential equations drawn from fluid dynamics" to describe human emotions. They also found "numerous fundamental conceptual and mathematical errors."
The part I found most telling sits in their walk through the 2004 paper. The model "predicted" 5.625 for high-performing teams against observed data of 5.614. It looks like a stunning match. Brown, Sokal and Friedman show the match came from an offset and a scaling constant applied to make the model output resemble the data. In engineering we call this overfitting. The model didn't predict anything. Somebody tuned it until it agreed.
Retraction Watch reported the 2005 paper was partially withdrawn, and the journal formally retracted the mathematical modeling. Wikipedia's summary notes Fredrickson conceded the math was "questionable." She still argues higher positivity ratios predict better outcomes, within bounds.
So the precise number is gone. What's left is a correlation from 60 teams in one corporate lab in the 1990s. Did the good teams perform well because they praised each other? Or did they praise each other because they were winning? The data doesn't tell you.
Why the Direction Still Holds
Here's where I side with Dan. Throw out the number and the underlying idea survives, because other research supports it.
In 2001 Roy Baumeister and colleagues reviewed a huge body of studies in a paper titled Bad Is Stronger Than Good. Their conclusion: bad emotions, bad parents, and bad feedback have more impact than good ones. Bad information gets processed more thoroughly. They wrote they found "hardly any exceptions."
If one critical comment weighs more than one kind one, you need more kind ones to keep the scales level. No retracted equations required.
The second piece of evidence matters more for how you give feedback. Avraham Kluger and Angelo DeNisi ran a 1996 meta-analysis of 607 effect sizes covering 23,663 observations. Feedback improved performance on average. But over a third of feedback interventions made performance worse.
The pattern they found: feedback works less well the more it pulls attention toward the person's self and away from the task.
Read those two findings together. Ratio counting misses the point. Five vague "great job" comments won't offset one remark about someone's competence. The direction of attention matters more than the tally.
What Counting Does to a Team
When a leader treats 5:1 as a quota, three things go wrong.
Praise turns into filler. You hunt for five nice things to say before you deliver the one thing you meant. People hear the setup and wait for the hit. The praise sandwich trains your team to ignore the bread.
Praise turns generic. Quotas reward volume. Volume rewards "nice work" and "great job, team." I've written about why generic team praise is noise and why equal praise isn't fair praise.
Criticism still lands on the person. The ratio says nothing about where the negative comment points. You hit your quota, then tell an engineer they "don't think things through." Kluger and DeNisi would predict a drop in performance, and five compliments won't save it.
Look at Your Code Reviews
Engineering teams run the clearest test of this every day. Pull requests.
Open the last ten reviews you left. Count the comments. Then count how many told the author something worked well and why.
In most engineering cultures I've seen, the answer is close to zero. Reviews exist to find problems, so reviewers write problems. A clean, clever refactor gets an "LGTM" and a merge. A missing null check gets three paragraphs.
Nobody designed it this way. It's the default. And the default teaches your engineers the only thing worth a reviewer's words is a mistake.
I'm not suggesting you pad reviews with compliments to hit a number. I'm suggesting the review is the best place to tell someone exactly what good looks like, because the code sits right there.
What I'd Do Instead of Counting
Make praise instructional
Specific praise teaches. "Nice work" teaches nothing. Compare it with this: "Splitting the migration into three reversible steps meant we rolled back in minutes instead of hours. Do it this way next time too."
Now the engineer knows what to repeat. So does everyone else reading the thread.
Point criticism at the work
Talk about the function, the test, the decision. Not the person's judgment, care, or ability. Kluger and DeNisi found feedback loses its effect as it moves toward the self.
Separate the two
Don't wrap criticism in praise to soften it. Give praise when you see the good thing. Give criticism when you see the problem. Your team will believe both.
Notice your silence
If you only speak up when something breaks, your silence becomes the loudest feedback you give. I wrote a whole post on what your silence teaches.
Why This Matters for Leaders
In my research, 99.5% of survey respondents said they've had one or more types of bad boss. A lot of those bosses weren't cruel. They were silent until something went wrong, then loud about it. Their teams learned to fear attention from the boss.
A ratio won't fix it. You fix it by noticing good work out loud, in detail, often enough for people to believe you mean it. If you want a structured look at how your team experiences your feedback, the Step It Up HR 360 tool is a good place to start.
Dan is right about the direction. Praise should outnumber criticism by a wide margin. Don't let a retracted equation tell you the margin is 2.9013, or 5.6, or 5. Let your team tell you.
When did you last tell an engineer exactly what they did right, in enough detail for them to do it again?